炼化装置中检测仪表与执行机构的异常诊断

Abnormal Diagnosis of Measuring Instrument and Actuators in Refining Plant

针对炼化装置控制回路中检测仪表与执行机构的异常诊断问题,提出一种相关分析法-聚类-支持向量描述(Support Vector Data Description,SVDD)方法。首先,选取控制回路中操纵变量与被控变量历史数据中具有动态变化特性的数据,通过相关分析法辨识其有限脉冲响应函数模型。接着,将多组有限脉冲响应模型参数构成一个数据矩阵,通过k-means算法进行聚类分析。然后,用聚类产生的类中数据较多的类训练SVDD超球体。最后,在控制回路中利用经过在线辨识出的有限脉冲响应模型参数与已经过训练的SVDD超球体进行异常诊断。利用某炼化厂实际生产数据,经过实验分析诊断出控制回路的异常原因,证实了该方法的有效性。

Considering abnormal diagnosis of the measuring instrument and actuators in the control circuit of refining and chemical units,a correlation analysis-clustering-support vector data description(SVDD)method was proposed to diagnose abnormal measuring instrument and actuators of the control loop.Firstly,having the data with dynamic characteristics in thehistorical data of manipulated variables and controlled variables in the control loop selected and the correlation analysis method adopted to identify the finite impulse response function model;then,having multiple groups of finiteimpulse response model parameters formed into a data matrix and k-means algorithm adopted to perform cluster analysis,including having the SVDD hypersphere trained with the class with more data among the classes generated by the clustering;finally,the finite impulse response model parameters identified online in the control loop and the obtained SVDD hypersphere were used for abnormal diagnosis.The experimental data in this paper comes from the actual productiondata of a temperature control loop in a domestic refinery and chemical plant,and the abnormal cause of the control loop was diagnosed through experimental analysis to prove the effectiveness of the above method.